[PATCH] libdrm: intel/Android.mk: Filter libdrm_intel library requirements on x86

[Emil Velikov]

On 30 January 2018 at 05:42, John Stultz <john.stultz at linaro.org> wrote:
> On Fri, Jan 26, 2018 at 10:33 AM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>> Hi all,
>>
>> Couple of ideas/notes,
>>
>> On 10 January 2018 at 20:36, Rob Herring <robh at kernel.org> wrote:
>>> On Wed, Jan 10, 2018 at 1:09 PM, John Stultz <john.stultz at linaro.org> wrote:
>>>> On Wed, Jan 10, 2018 at 5:48 AM, Rob Herring <robh at kernel.org> wrote:
>>>>> On Tue, Jan 9, 2018 at 11:25 PM, John Stultz <john.stultz at linaro.org> wrote:
>>>>>> When building AOSP after updating libdrm project to the
>>>>>> freedesktop/master branch, I've seen the following build errors:
>>>>>>
>>>>>> external/libdrm/intel/Android.mk: error: libdrm_intel
>>>>>
>>>>> This is only needed for i915 (not i965) now BTW. I'm not sure at what
>>>>> point we stop caring about i915.
>> If you're using any other Intel components - say Beignet or the va
>> driver, I think those still use libdrm_intel.
>>
>> An alternative solution IMHO, is to drop/tweak the API to not bother
>> libpciaccess.
>> I have some ancient cleanup/rework branch
>>
>> https://github.com/evelikov/libdrm/commits/intel-remove-legacy
>
> I'm not opposed to this, but I'm really not familiar with intel use
> cases and what would be ok or not there.
>
>
The unfortunate part is that people familiar don't have to
time/interest to weight in :-(
I might give it another try, one of these days. Unless someone beats me to it.

>>>>>> (SHARED_LIBRARIES android-arm64) missing libpciaccess
>>>>>> (SHARED_LIBRARIES android-arm64) You can set
>>>>>> ALLOW_MISSING_DEPENDENCIES=true in your environment if this is
>>>>>> intentional, but that may defer real problems until later in the
>>>>>> build.
>>>>>>
>>>>>> Using ALLOW_MISSING_DEPENDENCIES=true when building allows
>>>>>> things to function properly, but is not ideal.
>>>>>>
>>>>>> So basically, while I'm not including the libdrm_intel package
>>>>>> into the build, just the fact that the Android.mk file references
>>>>>> libpciaccess which isn't a repo included in AOSP causes the build
>>>>>> failure.
>>>>>>
>>>>>> So it seems we need some sort of conditional filter in the
>>>>>> Android.mk to skip over it if we're not building for intel.
>>>>>>
>>>>>> This is my initial attempt at solving this.
>>>>>>
>>>>>> Feedback would be greatly appreciated!
>>>>>>
>>>>>> I note that in the AOSP version of libdrm, the reference to
>>>>>> libpciaccess has been removed. See:
>>>>>>  https://android.googlesource.com/platform/external/libdrm/+/f6a1130dffae8de9ddd0c379066daf1df27fc8af%5E%21/
>>>>>> So I wonder if it make sense to instead remove this upstream as
>>>>>> well?
>>>>>
>>>>> Only if we drop i915.
>>>>
>>>> To be more precise, drop i915 for Android builds (I'm not suggesting
>>>> dropping it elsewhere, just for the Android.mk). I'm really not sure
>>>> which devices might be affected. Anyone able to point me to someone in
>>>> Intel who would know?
>>>
>>> The android-x86 folks would be the ones to ask. I added Chih-Wei.
>>>
>> A really silly question - how are you triggering any of this if you're
>> building on !x86?
>> Is that because the GPU driver is not selected thus you we fall-back
>> to "build all"?
>
> I think its mostly due the fact we're using the toplevel Android.mk
> which includes all Android.mk files in subdirectories.
>
That does not matter. Unless otherwise stated the objects are optional.
Thus they should not be build, unless...

Android changed the policy or you're somehow building stuff that's not required?

>> If so, it might be better to change things to:
>>  - error out if none selected
>>  - allow one to select "all", "x86", "arm" and similar groups thus
>> only the things that can build are build
>> eg. RobH had fun with x86 intrinsics while building the intel Vulkan
>> driver on ARM
>
> I'm not sure quite how to select a gpu driver with the Android build
> system, other then specifying it via a build variable, which is in
> effect what I'm trying to do with this patch.
>
> Other ideas?
>
Based on your input seems like it's not set (grep for
BOARD_GPU_DRIVERS), which results in "everything" being build.
Thus optional libraries (like libdrm_intel) are pulled causing the problem.

My earlier suggestion doesn't sound too crazy, although I'd check with
RobH and the Android-x86 people.
It might require one line change to the device manifest ;-)

Thanks
Emil